Demonstration of control and processing software for the ASKAP telescope

Dr Matthew Whiting1

1Csiro, Epping, Australia

 

In this presentation we demonstrate various aspects of the software systems used for the Australian Square Kilometre Array Pathfinder (ASKAP) telescope. This telescope is characterised by very large data rates, by virtue of its novel phased-array feed receivers, and it required specific solutions for the operation, control & monitoring, and data processing that were able to handle the large numbers of hardware elements and large data volumes. We will demonstrate software elements used to prepare and execute observations, and monitor their progress. The processing of observations on the galaxy supercomputer at the Pawsey Supercomputing Centre will be demonstrated, along with examples of the data products produced. And finally the CSIRO ASKAP Science Data Archive (CASDA), the repository of ASKAP observations and the go-to place for scientists to access data products will be demonstrated.


Biography:

Matthew Whiting is acting Group Leader for ATNF Science, part of CSIRO Astronomy & Space Science, leading a group responsible for research and science operations within the Australia Telescope National Facility. He has a PhD in astrophysics from University of Melbourne, and a career of working in astronomy and astronomical computing. Matthew leads the development of the processing pipelines for CSIRO’s ASKAP telescope, and is a member of the small software team responsible for the high-performance calibration & imaging software for ASKAP. He is also the sole developer of the widely-used Duchamp source-finding software, designed for 3D astronomical spectral-line surveys. He works closely with the Pawsey Supercomputing Centre, and is a member of Astronomy Australia Limited’s Astronomy e-Research Advisory Committee.

Indices of variability and their links to Australian rainfall in climate forecasts

Dr Ian Watterson1, Lauren Stevens1, Mark Collier1, Terry O’Kane2

1CSIRO, Aspendale, Australia,

2CSIRO, Hobart, Australia

 

CSIRO’s new Decadal Forecasting Project aims to investigate climate predictability and deliver multi-year climate forecasts for Australia. A prototype Climate Analysis and Forecast Ensemble System has been based on the ACCESS Ocean – GFDL AM2.1 coupled model. Given the potential importance of ENSO and the southern annular mode (SAM) to the variability of Australian rainfall, the links between these are explored first in a 100-year climatological simulation. Regression fields show that anomalies in atmospheric circulation linked to both the NINO34 ENSO index and a height-based SAM index extend through much of the southern hemisphere. Correlations between NINO34 and rainfall in eastern and northern Australia exceed –0.5 in summer and spring, in part through a link between NINO34 and SAM. These links can be seen also in two versions of ensemble forecast simulations, initialised from 1 January in years 2004-2016. Variability is enhanced for a year or more in the version that includes assimilation of observed ocean temperatures in the initial states and the use of mode-like perturbations. Given the links, predictive skill in forecasting NINO34 would imply some predictability in anomalies of SAM and Australian rainfall.


Biography:

Dr Ian Watterson is a senior principal research scientist at CSIRO Oceans and Atmosphere, located at Aspendale where he has worked since 1989. From a background in atmospheric dynamics, he has contributed to CSIRO climate modelling and climate impact programs over the years. He was one of the lead authors of the recent Australian projections, and also of the IPCC’s Climate Change 2007, and has written over 60 journal papers. Dr Watterson is currently contributing to CSIRO’s new Decadal Forecasting Project, along with co-authors of the submission to C3DIS.

Designing regional climate models for convective permitting resolution

Dr Marcus Thatcher1

1CSIRO Oceans And Atmosphere, Aspendale, Australia

 

Regional climate models are required to simulate atmospheric processes at spatial resolutions of kilometres and temporal resolutions of less than a minute for time periods of 100+ years.  Consequently, such models need to make extensive use of high performance computing facilities.  This presentation describes the development of CSIRO’s Conformal Cubic Atmospheric Model (CCAM), which employs a variable resolution, global cubic grid that can be focused over a region using a Schmidt transformation.  Since CCAM has no lateral boundaries, we use a scale-selective filter to assimilate global atmospheric circulation into the regional simulation.  This approach allows us to scale the simulation size depending on the available computing resources, step-down to finer resolutions much faster than for limited area approaches, as well as maintain the coupling between the regional and global scales.  CCAM employs MPI parallelisation, including the use of shared memory within a node, as well as an optional OpenMP parallelisation that improves the performance on Xeon Phi Knights Landing.  CCAM also employs a system for reading and writing model data in parallel, based on a clustering of compressed NetCDF4 files for each node.  These enhancements allow CCAM simulations to successfully scale beyond 20,000 cores and can achieve 2+ simulation years per day at 2 km resolution.  The resultant convective permitting simulations show improvements in the simulated rainfall (particularly for extreme rainfall), as well as for urban areas, coastal regions and complex orography.  In turn, the model has applications for urban flooding, air quality and extreme weather research.


Biography:

Dr Marcus Thatcher has been developing regional climate models at CSIRO for 14 years, and currently leads the Extreme Weather and Climate Team at the CSIRO Oceans and Atmosphere Climate Science Centre.

Deep Learning – A new approach for multi-label scene classification in Planetscope and Sentinel-2 imagery

Dr Yuri Shendryk1, Yannik Rist1, Dr Catherine Ticehurst2, Dr Peter Thorburn1

1CSIRO, Brisbane, Australia,

2CSIRO, Canberra, Australia

 

Motivated by the increasing availability of high-resolution satellite imagery, we developed deep learning models able to efficiently and accurately classify the atmospheric conditions and dominant classes of land cover/land use in commercial PlanetScope imagery acquired over the Amazon rainforest. In specific, we trained deep convolutional neural network (CNN) to perform multi-label scene classification of high-resolution (<10 m) satellite imagery. We also discuss the challenges and opportunities in training ensemble CNN models for multi-label scene classification. Our best performing model achieved an F-beta score of 0.91, which was only 2% short of the top performing model in the Understanding the Amazon from Space Kaggle competition. Finally, we investigate the transferability of our PlanetScope-trained models to freely available Sentinel-2 imagery acquired over the wet tropics of Australia. The models trained on PlanetScope imagery performed well when applied to Sentinel-2 imagery with F-beta score of 0.79. Similarly, the models trained on Sentinel-2 imagery achieved F-beta score of 0.80 when applied to PlanetScope imagery. This suggests that our CNN models are suitable for classifying the atmospheric conditions and dominant classes of land cover/land use in satellite imagery of similar resolution to that of PlanetScope and Sentinel-2 (i.e.  <10 m).


Biography:

Dr Yuri Shendryk is a Postdoctoral Fellow at CSIRO specializing in developing algorithms to process terabytes of satellite and airborne data. After earning a master’s degree in geophysics, he spent more than three years working and studying geospatial engineering in Sweden and Germany. In 2017 he received a PhD from UNSW, and his current research in CSIRO is centred around the integration of remote sensing and machine learning for forest health monitoring and precision agriculture.

Computational Chemistry and Distributed Computing

Dr Manolo Per1, Dr Deidre Cleland1

1Data61 CSIRO, Docklands, Australia

 

Accurate prediction of the quantum-scale properties of materials and chemical processes requires a significant amount of computational resources. The methods commonly used need large amounts of memory and fast network interconnects, and so depend on the use of supercomputers for practical applications. As a result, a considerable fraction of the world’s supercomputer time is consumed by these calculations.

However, these methods are unable to make use of the very large numbers of compute cores which are common in contemporary supercomputers. More important issues arise as we head toward the Exascale era, where the emergence of architectures using heterogeneous solutions such as accelerators present additional challenges to both theory and software implementation.

In this talk I will describe our approach to solving these problems, which involves a very specific implementation of a stochastic power method for solving the quantum mechanical Schrodinger equation. The resulting algorithm provides almost perfect scalability, enables the use of heterogeneous environments, is implicitly fault tolerant, and can be used in truly distributed environments with very slow network connections.

These developments give the unique ability to exploit virtually any computational resource, including current and future supercomputers, idle PCs, and commercial clouds


Biography:

Dr. Manolo Per leads Data61’s Molecular and Materials Modelling Team, an inter-disciplinary team of computational scientists who develop and implement new techniques in fundamental and data-driven chemical and materials science.

Manolo’s background is in electronic-structure theory, and his main research interests are in the development of methods for solving the quantum many-body problem, and their practical implementation in computer software.

Advances and challenges in simulating the Great Barrier Reef environment with a fine-resolution near-real-time modelling system

Dr Nugzar Margvelashvili1, Dr John  Andrewartha1, Dr Mark Baird1, Dr Mike  Herzfeld1, Dr Jonathan  Hodge2, Dr Emlyn  Jones1, Dr Mathieu  Mongin1, Dr Farhan  Rizwi1, Dr Barbara   Robson3,4, Dr Jenny  Skerratt1, Dr Karen  Wild-Allen1, Dr Monika  Wozniak1

1CSIRO Oceans and Atmosphere, Hobart, Australia,

2CSIRO Oceans and Atmosphere, Brisbane, Australia,

3CSIRO Land and Water, Canberra, Australia,

4Australian Institute of Marine Science, Townsville, Australia

 

This talk summarises recent advances and challenges in developing and maintaining a 3D fine resolution modelling system simulating water quality in the Great Barrier Reef (GBR) region. This modelling system was developed through the multi-year multi-institutional eReefs project and simulates hydrodynamics, sediment transport, cycling of nutrients and optical properties of water masses on the shelf in the near-real-time regime. Environmental processes on the GBR shelf are characterised by an extremely wide range of spatial and temporal scales. To traverse these scales, a capability has been developed for nesting a high resolution RElocatable Coastal Ocean Model (RECOM) within a large-scale regional model via a web interface.

A number of strategies have been used to constrain this modelling system with observations, including an emulator-assisted calibration of sediment model parameters and ensemble-based assimilation of remotes sensing and ground measurements into a complex biogeochemical model. The quality of the calibrated models varies across both space and time. The model generates large volumes of gridded data representing more than 100 prognostic and diagnostic state variables daily (and hourly) over the 3D grid with 47 layers in vertical, and 180 x 600 grid cell in horizontal directions. Processing and analysing such volumes of data is a challenge, particularly for practitioners with a limited IT background. Current practices facilitating uptake of knowledge from such data will be outlined and future strategies will be discussed.


Biography:

Nugzar Margvelashvili is a coastal sediment transport modeller, currently with CSIRO Marine and Atmospheric Division, Hobart.

Workspace in an Industry 4.0 Environment

Damien Watkins1

1CSIRO, Clayton, Australia

 

Workspace is a cross-platform, domain agnostic scientific workflow and application development platform. Developed by CSIRO and in use for over 10 years, Workspace aims to shorten the path from research to impact.

Workspace is currently a key component of a number of research projects looking into manufacturing process automation. This presentation describes key insights from this work, demonstrating how Workspace can be used to maximise both software reuse and setup flexibility in an industry 4.0 setting.


Biography:

Dr Damien Watkins is the Research Team Lead for the Computational Software Engineering and Visualisation Team at CSIRO/ Data61. His team is responsible for the development of 1) Workspace, a scientific workflow platform used on projects across CSIRO as well as outside CSIRO, 2) a number of Workspace-based applications and 3) the Software Product Platform (SPP). Workspace provides a framework that allows researchers to focus on their science, to develop robust and sustainable software and to accelerate software development timeframes with a faster path to market for their commercialisation. The SPP is a suite of higher level, re-usable Workspace based components for performing key functions (e.g. meshing, visualisation, licensing, etc.) whose availability can provide a second level of cost savings and commercialisation acceleration. Workspace has been available for external usage since 2014.

Clustering of Electricity Customers based on Load Characteristics

Dr Omid R.E. Motlagh1, Dr Adam Berry, Mr Lachlan O’Neil

1CSIRO-Energy , Mayfield West, Australia

 

 

To better understand the patterns of residential electricity load behaviour, it is often useful to group similarly behaving homes together.  Most contemporary clustering methods that would otherwise be well suited to the task, however, perform poorly on the type of noisy and patchy time-series datasets common to the residential energy domain.  They are also limited by a strong preference for time-series data that is synchronised: of both the same extent and sampling frequency.  We describe a model-based load-clustering method to address the problems of unequal and asynchronous time series as well as lack of robustness against input noise. The clustering results – using a large dataset of 7000 homes – show around 14% improvement of coincidence factor averaged across the obtained clusters versus the full set of the homes. The method also results in around 73% similarity between clusters using synchronous and asynchronous-unequal profiles. It also shows high robustness of more than 90% against a heavy mixed noise with signal-to-noise ratio as low as 29%.  There are many applications of this load clustering technique including handling incomplete and noisy datasets, some of which are exemplified in this presentation.


Biography:

Omid Motlagh is a research scientist with the CSIRO-Energy based in Newcastle. His project-research interest is Energy, Urban and Systems Engineering, Machine Learning, and Experimental Mathematics. Adam Berry is the Grids and Energy Efficiency Program Research Group Leader and is the director of the Energy Use Data Model project for the Department of the Environment and Energy. Lachlan O’Neil is an Energy Data Statistician and Scientist for the Energy Use Data Model project for the Department of Environment and Energy.

Automated well log interpretation for seismic inversion

Dr Christopher Dyt1, Dr Irina Emeylanova1, Dr Ben Clennell1

1CSIRO Energy, Kensington, Australia

 

To perform robust seismic inversions, accurate velocity and impedance profiles need to be defined for each of the different rock types present in the subsurface, both reservoir lithologies (e.g. clean sandstone, porous limestone) and the overburden rocks (shales, marls). In order to generate these profiles rapidly, objectively and in a repeatable manner we have developed software to automatically determine lithology from well logs.

To classify the rock sequence into types relevant to seismic inversion we must first capture an expert’s methodology of interpreting facies from the well logs, and convert these into mathematical algorithms. These algorithms first of all help us identify rapidly any non siliciclastic rocks (such as limestone, coal and pyrite-rich sediments) due to their very distinctive signals. The remainder of the well log section is then grouped using a k-mean clustering technique using the basic petrophysical well log properties such as gamma ray, density, neutron porosity and photoelectric factor that a geologist uses to classify the section. These clusters are then classified, according to the expert’s rules, into broad textural categories: sand, coarse to medium grain, coarse to fine grain, and shale.

The technique was applied to four wells in the Carnarvon Basin, North West Shelf, Western Australia. Despite the wells being hundreds of kilometres apart, a remarkably consistent depth v velocity and depth v impedance profile was obtained for the different rock types. These common trends should help to establish reliable prior probabilities for velocity and impedance that can be used to improve seismic inversion workflows.


Biography:

Chris Dyt is an applied mathematician with over 20 years experience in developing numerical solutions in the oil and gas sector. Chris has had the good fortune to work across a broad array of projects, including sequence stratigraphy, pipe line flow, well injection prediction and fracture modelling. In the last two years, Chris has joined the Data Analytics Group based at ARRC WA.

Massively parallel quantum scale simulations of biological systems

Dr Deidre M. Cleland1, Miss Emily Fletcher1, Dr Manolo C. Per1

1Data61 CSIRO, Docklands, Australia

 

Introduction

The properties and behaviour of biologically important molecules are highly dependent on the correlated behaviour of their constituent electrons.  An accurate description of these quantum effects is therefore crucial for reliable molecular simulations.  However, computational modelling of electron correlation is very computationally demanding.  To take full advantage of existing and future computational resources, we require accurate modelling methods that will scale efficiently to enormous numbers of computing cores.  The most practical way to achieve this is through stochastic techniques.

Method and Results

Quantum Monte Carlo (QMC) methods are a set of stochastic electronic structure techniques that have been shown to accurately describe a range of molecular systems, with near perfect scaling to thousands of high performance computing cores.  However, the computational cost of calculations can still be significant.  In this poster, we describe the development of the QMC algorithm towards a version that can be applied to larger biological systems, with a particular focus on DNA, and the goal of modelling the series of base pairs that make up a twist of the double helix structure.  For this, the QMC algorithm is modified to promote size consistency, allowing energy differences to be calculated with substantially reduced computational cost.  Additional developments focus on ensuring the algorithm scales efficiently to even larger number of parallel processors, with the goal of simulating large molecules in the cloud.

Conclusion

Through algorithmic developments to promote size consistency, and greater efficiency for parallelisability, QMC is developed for accurate modelling of biologically important molecules.


Biography:

Dr Deidre Cleland is a Research Scientist in the Molecular and Materials Modelling group at Data61.  Dr Cleland was awarded her PhD in 2012, for development of the initiator Full Configuration Interaction Quantum Monte Carlo method, in collaboration with her colleagues in the Alavi research group at the University of Cambridge.

Upon completion of her PhD, Dr Cleland spent 18 months working as a Data Analyst in London before joining CSIRO in 2014. Her research at Data61 is primarily focused on cheminformatics and Quantum Monte Carlo methods for accurate quantum-scale simulations of molecules.

ABOUT AeRO

AeRO is the industry association focused on eResearch in Australasia. We play a critical coordination role for our members, who are actively transforming research via Information Technology. Organisations join AeRO to advance their own capabilities and services, to collaborate and to network with peers. AeRO believes researchers and the sector significantly benefit from greater communication, coordination and sharing among the increasingly different and evolving service providers.

Conference Managers

Please contact the team at Conference Design with any questions regarding the conference.
© 2019 Conference Design Pty Ltd